ML-DSA AArch64/32: assembly implementation

.github/configs/pq-all.json

@@ -189,6 +189,46 @@
  "configure": ["--host=aarch64-linux-gnu", "--enable-cryptonly",
   "--enable-armasm=inline", "--enable-experimental",
   "--enable-frodokem=aes,ephemeral,sve,sme", "--disable-examples"]},
+{"name": "mldsa-mlkem-arm32-armasm", "minutes": 0.5, "check": false,
+ "cc": "ccache arm-linux-gnueabihf-gcc",
+ "comment": "AArch32 NEON ML-DSA + ML-KEM and the batched Keccak they share. Runs the KATs under qemu: these are wrong-answer risks, so a build-only job would not catch them.",
+ "configure": ["--host=arm-linux-gnueabihf", "--enable-cryptonly",
+  "--enable-armasm", "--enable-mldsa", "--enable-mlkem",
+  "--disable-examples"],
+ "run": [["env", "QEMU_LD_PREFIX=/usr/arm-linux-gnueabihf",
+   "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "mldsa-mlkem-arm32-armasm-inline", "minutes": 0.5, "check": false,
+ "cc": "ccache arm-linux-gnueabihf-gcc",
+ "comment": "Same, through the inline-asm backend. It is a separate code path from the .S - operand register allocation differs - and has carried defects the .S build did not.",
+ "configure": ["--host=arm-linux-gnueabihf", "--enable-cryptonly",
+  "--enable-armasm=inline", "--enable-mldsa", "--enable-mlkem",
+  "--disable-examples"],
+ "run": [["env", "QEMU_LD_PREFIX=/usr/arm-linux-gnueabihf",
+   "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "mldsa-mlkem-arm32-armasm-no-neon", "minutes": 0.5, "check": false,
+ "cc": "ccache arm-linux-gnueabihf-gcc",
+ "comment": "WOLFSSL_ARMASM_NO_NEON: scalar SHA-3 assembly with the C ML-DSA, and the only AArch32 job covering the non-NEON dispatch.",
+ "configure": ["--host=arm-linux-gnueabihf", "--enable-cryptonly",
+  "--enable-armasm", "--enable-mldsa", "--enable-mlkem",
+  "--disable-examples", "CPPFLAGS=-DWOLFSSL_ARMASM_NO_NEON"],
+ "run": [["env", "QEMU_LD_PREFIX=/usr/arm-linux-gnueabihf",
+   "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "mldsa-mlkem-aarch64-armasm", "minutes": 0.5, "check": false,
+ "cc": "ccache aarch64-linux-gnu-gcc",
+ "comment": "AArch64 NEON ML-DSA + ML-KEM. QEMU_CPU=max so the FEAT_RDM and FEAT_SHA3 dispatchers take their instruction-extension paths.",
+ "configure": ["--host=aarch64-linux-gnu", "--enable-cryptonly",
+  "--enable-armasm", "--enable-mldsa", "--enable-mlkem",
+  "--disable-examples"],
+ "run": [["env", "QEMU_LD_PREFIX=/usr/aarch64-linux-gnu", "QEMU_CPU=max",
+   "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "mldsa-mlkem-aarch64-armasm-inline", "minutes": 0.5, "check": false,
+ "cc": "ccache aarch64-linux-gnu-gcc",
+ "comment": "Same, through the inline-asm backend.",
+ "configure": ["--host=aarch64-linux-gnu", "--enable-cryptonly",
+  "--enable-armasm=inline", "--enable-mldsa", "--enable-mlkem",
+  "--disable-examples"],
+ "run": [["env", "QEMU_LD_PREFIX=/usr/aarch64-linux-gnu", "QEMU_CPU=max",
+   "./wolfcrypt/test/testwolfcrypt"]]},
 {"name": "mlkem-old-ids", "minutes": 0.4,
  "configure": ["--enable-intelasm", "--enable-sp-asm",
   "--enable-mlkem=yes,kyber,ml-kem",

.github/workflows/pq-all.yml

@@ -1,3 +1,3 @@
 name: Quantum Resistant Tests
 
 # START OF COMMON SECTION
@@ -49,7 +49,7 @@
       - name: Install dependencies
         uses: ./.github/actions/install-apt-deps
         with:
-          packages: autoconf automake libtool build-essential bubblewrap crossbuild-essential-arm64 crossbuild-essential-armhf
+          packages: autoconf automake libtool build-essential bubblewrap crossbuild-essential-arm64 crossbuild-essential-armhf qemu-user
           ghcr-debs-tag: ubuntu-24.04-minimal
 
       # ccache via the cross-platform composite; the script passes the

src/include.am

@@ -1351,6 +1351,22 @@ if BUILD_INTELASM
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/wc_mldsa_asm.S
 endif BUILD_INTELASM
 endif !BUILD_X86_ASM
+if BUILD_ARMASM
+if !BUILD_ARM_THUMB
+if BUILD_ARMASM_INLINE
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-mldsa-asm_c.c
+else
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-mldsa-asm.S
+endif !BUILD_ARMASM_INLINE
+endif !BUILD_ARM_THUMB
+endif BUILD_ARMASM
+if BUILD_ARMASM_NEON
+if BUILD_ARMASM_INLINE
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-mldsa-asm_c.c
+else
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-mldsa-asm.S
+endif !BUILD_ARMASM_INLINE
+endif BUILD_ARMASM_NEON
 endif
 
 if BUILD_WC_LMS
@@ -2152,6 +2168,22 @@ if BUILD_INTELASM
 src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/wc_mldsa_asm.S
 endif BUILD_INTELASM
 endif !BUILD_X86_ASM
+if BUILD_ARMASM
+if !BUILD_ARM_THUMB
+if BUILD_ARMASM_INLINE
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-mldsa-asm_c.c
+else
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-32-mldsa-asm.S
+endif !BUILD_ARMASM_INLINE
+endif !BUILD_ARM_THUMB
+endif BUILD_ARMASM
+if BUILD_ARMASM_NEON
+if BUILD_ARMASM_INLINE
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-mldsa-asm_c.c
+else
+src_libwolfssl@LIBSUFFIX@_la_SOURCES += wolfcrypt/src/port/arm/armv8-mldsa-asm.S
+endif !BUILD_ARMASM_INLINE
+endif BUILD_ARMASM_NEON
 endif
 
 if BUILD_WC_LMS

wolfcrypt/src/include.am

@@ -32,6 +32,7 @@ EXTRA_DIST += wolfcrypt/src/wc_mldsa_asm.asm
 EXTRA_DIST += wolfcrypt/src/port/arm/armv8-aes-asm.asm
 EXTRA_DIST += wolfcrypt/src/port/arm/armv8-chacha-asm.asm
 EXTRA_DIST += wolfcrypt/src/port/arm/armv8-curve25519.asm
+EXTRA_DIST += wolfcrypt/src/port/arm/armv8-mldsa-asm.asm
 EXTRA_DIST += wolfcrypt/src/port/arm/armv8-mlkem-asm.asm
 EXTRA_DIST += wolfcrypt/src/port/arm/armv8-poly1305-asm.asm
 EXTRA_DIST += wolfcrypt/src/port/arm/armv8-sha256-asm.asm